Scientific Computing I
Module 5: Heat Transfer Discrete and Continuous Models
Michael Bader
Lehrstuhl Informatik V
Winter 2016/2017
Part I: Discrete Models
Motivation: Heat Transfer
Wiremesh Model
A Finite Volume Model
Time Dependent Model
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
Part I
Discrete Models
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
Motivation: Heat Transfer
objective: compute the temperature distribution of some object
under certain prerequisites:
temperature at object boundaries given
heat sources
material parameters
observation from physical experiments:
q k T
heat flux proportional to temperature differences
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
A Wiremesh Model
consider rectangular plate as fine mesh of wires
compute temperature xi,j at nodes of the mesh
x i,j+1
x i1,j x i,j x i+1,j
x i,j1
hy
hx
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
A Wiremesh Model (2)
model assumption:
temperatures in equilibrium at every mesh node
equilibrium: steady state (of temperature), energy balance (inflow =
outflow) in each node of the mesh
incoming temperature fluxes at point i,j via the four wires:
from the left: k xi1,j xi,j
from the right: k xi+1,j xi,j
from below: k xi,j1 xi,j
from above: k xi,j+1 xi,j
equation for steady state: sum over all fluxes = zero:
xi,j =
1
xi1,j + xi+1,j + xi,j1 + xi,j+1
4
for all temperatures xi,j .
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
A Wiremesh Model (3)
temperature known at (part of) the boundary; for example
x0,j = Tj
models a heated/cooled wall with constant temperature Tj at the left
boundary.
temperature flux known at (part of) the boundary; for example
xi,0 = xi,1
xi,1 xi,0 = 0
models an isolated wall at the lower boundary.
heat sources: temperature given at a certain position i, j:
xi,j = Ts .
task: solve Linear System of Equations
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
A Finite Volume Model
object: a rectangular metal plate (again)
model as a collection of small connected rectangular cells
hy
hx
examine the heat flux across the cell edges
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
A Finite Volume Model (2)
model assumption: temperatures in equilibrium in every grid cell
heat flux across a given edge is proportional to
temperature difference (T1 T0 ) between the adjacent cells
length h of the edge
e.g.: heat flux across the left edge:
(left)
qi,j
= kx Ti,j Ti1,j hy
note: heat flux out of the cell (and kx > 0)
heat flux across all edges determines change of heat energy:
qij
= kx Tij Ti1,j hy + kx Tij Ti+1,j hy
+ ky Tij Ti,j1 hx + ky Tij Ti,j+1 hx
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
A Steady-State Model
heat sources: consider additional source term Fi,j due to
external heating
radiation
Fi,j = fi,j hx hy (fi,j heat flux per area)
equilibrium with source term requires qi,j + Fi,j = 0:
fi,j hx hy
= kx hy 2Ti,j Ti1,j Ti+1,j
ky hx 2Ti,j Ti,j1 Ti,j+1
again, Linear System of Equations
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
10
Towards a Time Dependent Model
idea: set up an ODE for each cell
simplification: no external heat sources or sinks, i.e. fi,j = 0
change of temperature per time is proportional to heat flux qi,j (t) into the
cell (no longer 0):
d
Ti,j (t) =
dt
=
+
c qi,j (t)
x
2Tij (t) + Ti1,j (t) + Ti+1,j (t)
hx
y
2Tij (t) + Ti,j1 (t) + Ti,j+1 (t)
hy
solve a system of ODEs
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
11
Boundary Conditions
(Finite Volume Models)
temperature known in boundary layer cells; for example
(left)
q1,j
= kx T1,j T0,j hy = kx T1,j T (x0,j ) hy
with T0,j = T (x0,j ) not an unknown!
(models a heated/cooled wall with constant temperature T (x0,j ) at the left
boundary)
(left)
temperature flux known in boundary layer cells; e.g. q1,j
= 0:
f1,j hx hy
= kx hy T1,j T2,j
ky hx 2T1,j T1,j1 T1,j+1
models an isolated wall at the left boundary.
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
12
Part II: A Continuous Model The Heat Equation
From Discrete to Continuous
Derivation of the Heat Equation
Variants of the Heat Equation
Boundary and Initial Conditions
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
13
Part II
A Continuous Model The Heat
Equation
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
14
From Discrete to Continuous
remember the discrete model:
fi,j
kx
2Ti,j Ti1,j Ti+1,j
hx
ky
2Ti,j Ti,j1 Ti,j+1
hy
assumption: heat flux across edges is proportional to temperature
difference
(left)
qi,j
= kx Ti,j Ti1,j hy
in reality: heat flux proportional to temperature gradient
(left)
khy
qi,j
Ti,j Ti1,j
hx
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
15
From Discrete to Continuous (2)
replace kx by k/hx , ky by k /hy , and get:
k
2Ti,j Ti1,j Ti+1,j
2
hx
k
2 2Ti,j Ti,j1 Ti,j+1
hy
fi,j
consider arbitrarily small cells: hx , hy 0:
fi,j = k
2T
x 2
k
i,j
2T
y 2
i,j
leads to a partial differential equation (PDE):
k
2 T (x, y ) 2 T (x, y )
+
x 2
y 2
= f (x, y )
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
16
Derivation of the Heat Equation
finite volume model, but with arbitrary control volume D
change of heat energy (per time) is a result of
transfer of heat energy across Ds surface,
heat sources and sinks in D (external influences)
resulting integral equation:
Z
cT dV =
Z
q dV +
kT ~n dS
density , specific heat c, and heat conductivity k are material parameters
heat sources and sinks are modelled in term q
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
17
Derivation of the Heat Equation (2)
according to theorem of Gauss:
k T ~n dS =
Z
kT dV
D
leads to integral equation for any domain D:
Z
cTt q k T dV = 0
D
hence, the integrand has to be identically 0:
Tt = T +
q
,
c
:=
k
c
> 0 is called the thermal diffusion coefficient (since the Laplace
operator models a (heat) diffusion process)
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
18
Variants of the Heat Equation
Different scenarios:
vanishing external influence, q = 0:
Tt = T
alternative notation
T
=
t
2T
2T
2T
+
+
x 2
y 2
z 2
equilibrium solution, Tt = 0:
0 = T +
q
c
T =
q
c
Poissons Equation
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
19
Boundary Conditions
Dirichlet boundary conditions:
fix T on (part of) the boundary
T (x, y , z) = (x, y , z)
Neumann boundary conditions:
fix T s normal derivative on (part of) the boundary:
T
(x, y , z) = (x, y , z)
n
special case: insulation
T
(x, y , z) = 0
n
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
20
Part III
Discretization: Finite Difference and
Finite Volume Methods
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
21
Part III: Discretization: Finite Difference and Finite
Volume Methods
From Continuous Back To Discrete Models
Finite Difference Methods
Meshes for Finite Difference Discretisation
Finite Difference Discretisation
Resulting Linear System of Equations
Discretisation Stencils
Finite Volume Methods
Finite Volume Meshes
Finite Volume Discretisation
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
22
From Continuous Back To Discrete Models
Continuous Models:
result from a limit process (h 0) from discrete model
(wire mesh, finite volume)
opposite route discretisation
Discretisation methods:
Finite Differences:
replace derivative by difference quotients
Finite Volumes:
compute fluxes on boundary of control volumes and examine
conservation laws
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
23
The Model Problem
2D Poisson Equation:
2
2
u(x,
y
)
+
u(x, y) = f (x, y )
x 2
y 2
on the unit square = (0, 1)2
with Dirichlet boundary conditions:
u(x, y) = g(x, y )
on
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
24
Meshes for Finite Difference Discretisation
regular, Cartesian mesh; analogous to the wire-mesh model:
x i,j+1
hy
x i1,j x i,j x i+1,j
x i,j1
hy
hx
hz
hx
compute approximate value of u for each mesh point:
uij u(xij )
uijk u(xijk )
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
25
Finite Difference Discretisation
replace partial derivative (at each mesh point) by difference quotient:
2u
(xi,j )
x 2
2u
(xi,j )
y 2
u(xi+1,j ) 2u(xi,j ) + u(xi1,j )
hx2
u(xi,j+1 ) 2u(xi,j ) + u(xi,j1 )
hy2
leads to Linear System of Equations (h := hx = hy ):
1
h2
ui+1,j + ui,j+1 4ui,j
f (xi,j ) xi,j (0, 1)2
u(xi,j )
g(xi,j ) xi,j
+ui,j1 + ui1,j
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
26
Resulting Linear System of Equations
matrix-vector notation of the system:
Ah xh = fh
xh a vector of all unknowns uij
requires numbering of the unknowns
using row-wise numbering, e.g.:
xh = (u1,1 , . . . , u1,n , u2,1 , . . . , u2,n , . . . , un,1 , . . . , un,n )
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
27
Resulting Linear System of Equations (2)
Ah is a sparse matrix (only 5 diagonals are non-zero)
Ah is block-tridiagonal:
Ah =
1
h2
Bh
I
..
.
..
..
..
I
Bh
Bh = tridiag(1, 4, 1), where I is the unit matrix
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
28
Notation: Discretisation Stencils
idea: illustrate the matrix structure via a so-called discretisation stencil
represents one row of the matrix
matrix elements ordered according to their geometrical orientation
discretisation stencil for Poisson equations:
1D:
1
1 2
h2
2D:
1
1
h2
1
4
1
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
29
Finite Volume Methods Meshes
domain subdivided into grid cells/elements ij :
hy
hx
consider cell averages uij for each cell ij , i.e., u(x, y ) uij
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
30
Finite Volume Discretisation
integrate over grid cells ij :
2u
2u
(x,
y)
+
(x, y ) dx dy =
x 2
y 2
Z
ij
Z
f (x, y ) dx dy
ij
|
integration by parts:
2u
(x, y ) dx dy
x 2
yj+ 1
ij
Z
=
x 1
i+ 2
u
dy
(x, y)
x
x 1
i
yj 1
ij
{z
}
:= fij hx hy
2u
(x, y ) dx dy
y 2
xi+ 1
2
Z
=
xi 1
2
y 1
j+ 2
u
dx
(x, y)
y
y 1
j
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
31
Finite Volume Discretisation (2)
remember: u(x, y) = uij in each ij
thus approximation of derivatives on edges:
ui+1,j uij
u
x x 1
hx
i+
2
ui,j+1 uij
u
y y 1
hy
j+
uij ui1,j
u
x x 1
hx
i
2
uij ui,j1
u
y y 1
hy
j
again leads to Linear System of Equations:
1
hx
ui+1,j 2ui,j + ui1,j hy +
1
hy
ui,j+1 2ui,j + ui,j1 hx = fij hx hy
More detailed computation for first term:
y
Zj+ 2
j 1
2
x 1
Zj+ 2
i+ 2
ui+1,j uij
uij ui1,j
hy
u
(x, y)
dy =
dy =
ui+1,j 2uij + ui1,j
x
h
h
h
x
x
x
x 1
i
j 1
2
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
32
Finite Volume Discretisation More General . . .
typical formulation for first-order PDEs:
u
F (u(x, y)) G(u(x, y))
+
+
dx dy =
t
x
y
...
ij
and analogously:
yj+ 1
F (u(x, y))
dx dy =
x
xi 1
yj 1
ij
xi+ 1
G(u(x, y))
dx dy =
y
xi 1
ij
for Poisson Equation: F (u) =
Z
2
x u,
x 1
i+ 2
F (u(x, y ))
dy
y 1
j+ 2
G(u(x, y))
dx
yj 1
2
etc.
Michael Bader | Scientific Computing I | Heat Transfer Discr. & Cont. Models | Winter 2016/2017
33
